Meta launches open source Llama 3.3, shrinking powerful bigger model into smaller size

The 70B-Llama 3.3 is specifically optimized for cost-effective inference, with token generation costs as low as $0.01 per million tokens.